The CEO of Carta Healthcare discusses recent survey results that suggest why successful pilots often stall, why EHR integration has overtaken trust as the leading adoption barrier – and why more clinical leaders are taking ownership of AI strategy.
Artificial intelligence has produced no shortage of successful healthcare pilot projects. Yet many provider organizations still struggle to expand those early wins into enterprise-wide capabilities.
A new survey suggests the question is no longer whether AI can deliver value. Instead, hospitals increasingly are wrestling with the operational realities of integration, governance, ownership and workflow that determine whether AI becomes part of everyday clinical practice or remains trapped in isolated demonstrations.
Brent Dover is CEO of Carta Healthcare, which organized that survey. The company automates and streamlines clinical data management using AI combined with human clinical experts. He said the research reveals an industry that has largely gained confidence in AI’s potential – but still faces organizational barriers that technology alone cannot solve.
Not expanding at pace
The new survey found that 71% of healthcare organizations reporting measurable value from AI still are not expanding those initiatives at pace. That statistic changes the conversation, Dover says.
“The most useful thing about this finding is what it rules out,” he said. “The pilots proved that it does. What stalls is everything that comes after the proof.”
According to Dover, pilot projects succeed because they are intentionally narrow. They typically have a committed champion, a clearly defined workflow and concentrated organizational attention. Once organizations attempt to spread those projects across departments, however, those favorable conditions disappear.
“Success in a controlled setting tells you the model is capable,” he said. “It does not tell you the organization is ready to operationalize it.”
Instead, health systems encounter competing priorities, budget pressures, inconsistent workflows and unclear accountability. Those, he argued, are operational challenges rather than AI shortcomings.
Integration moves to the forefront
The survey found that EHR integration difficulty was the leading barrier to AI adoption, cited by 44% of respondents. That placed it well ahead of clinician trust and regulatory concerns, each identified by 26%.
Dover believes the finding reflects where AI ultimately succeeds or fails: inside clinicians’ existing workflows.
“A tool can be accurate in a demo and still fail the moment a clinician has to leave their workflow to use it,” he said. “Anything that sits outside it adds steps, and added steps are where adoption quietly dies.”
He noted that integration extends beyond technical interfaces. Clinical documentation is often inconsistent and highly contextual. Extracting meaningful information requires understanding clinical intent rather than simply moving data between systems.
For CIOs, CMIOs and digital health leaders, Dover recommends treating integration as a core purchasing requirement instead of something to address after signing a contract.
“Ask vendors to prove integration in your environment, with your data and your workflows, before you commit,” he said.
He also encourages health systems to evaluate who carries the operational burden. Organizations should favor vendors capable of delivering trustworthy information directly into existing workflows rather than requiring internal teams to absorb additional complexity.
Clinical ownership becomes essential
Another notable survey finding is that clinical leaders now most frequently own AI strategy, surpassing both IT organizations and executive leadership. At the same time, 26% of respondents reported having no clearly defined AI owner.
Dover views those results as two sides of the same story. “When clinical leadership is the most common owner of AI strategy, it means these are no longer viewed as technology experiments,” he said. “They are clinical decisions with clinical consequences.”
The absence of clear ownership concerns him even more because promising projects frequently lose momentum once the excitement surrounding an initial pilot fades.
“One in four organizations is trying to adopt AI with no one accountable for the strategy,” he said. “A pilot without an owner has nowhere to go once the initial enthusiasm fades.”
He advocates a governance model anchored by clinical leadership, with IT responsible for integration, security and infrastructure while executive leadership provides funding and oversight. Most importantly, one clinical leader should remain accountable for outcomes.
Vendor evaluation
Perhaps the survey’s strongest consensus involved vendor evaluation. Ninety-two percent of respondents said deep clinical domain expertise is critical when assessing an AI vendor.
Dover said the result reflects lessons organizations have learned through experience. “Generic AI, however capable in the abstract, does not on its own know the difference between a case that is documented cleanly and one that is clinically ambiguous,” he said.
Rather than focusing primarily on product demonstrations, he believes health systems should examine whether vendors understand clinical workflows, can demonstrate measurable outcomes at peer organizations and are willing to share performance risk.
“The investments that do scale are evaluated on different questions,” he said. “Does the vendor understand the clinical work well enough to be trusted with it? Will they show peer outcomes rather than promises?”
He added that healthcare organizations increasingly expect evidence that vendors can consistently deliver results under real operational conditions rather than relying on successful pilot projects alone.
From proof of concept to enterprise capability
Taken together, the survey findings suggest healthcare is entering a new phase of AI adoption. Questions about whether AI can produce value are gradually giving way to questions about implementation, governance and operational sustainability.
Hospitals that successfully scale AI appear likely to be those that establish clear ownership, insist on seamless workflow integration, evaluate vendors on demonstrated outcomes and prioritize clinical expertise throughout implementation, he said.
For healthcare technology executives, the survey points toward a practical roadmap. Enterprise AI success depends not simply on deploying increasingly sophisticated models but on embedding those capabilities into everyday clinical operations in ways clinicians trust and use.
If organizations can solve those operational challenges, Dover believes AI has the potential to move beyond isolated pilots and become a durable enterprise capability that consistently improves both clinical and operational performance.
Disciplined AI investing
Dover said the survey also indicates that healthcare buyers have become more disciplined in how they evaluate AI investments. Rather than being persuaded by technical capability alone, leaders increasingly want measurable evidence that an implementation will fit existing workflows, reduce manual effort and continue producing reliable results after the initial deployment period.
He believes that represents a healthy evolution for the industry because it shifts attention away from short-term excitement and toward long-term operational performance.
In his view, organizations that combine clinical leadership, realistic governance, careful integration planning and outcome-based vendor evaluation will be positioned to expand AI more confidently across the enterprise while avoiding the cycle of promising pilots that never become routine practice.
